- I've made a 7B model run on a laptop GPU that had no business running one.SNKT
- I've rebuilt RAG from open-source parts — embeddings, chunking, rerankers, quantization.FWUMP
- I've put guardrails on a medical AI so it never plays doctor.NOPE
- I've reconciled five marketplaces before breakfast.SYNC
- I've benchmarked FAISS, Milvus and Chroma until retrieval actually held up.HMMM
- I run the weekly disaster-recovery backups. Nobody writes comics about backups....
- Zero data-loss incidents. Not once.KNOCK
ALRIGHT. LET'S DO THIS ONE LAST TIME.
KIRTAN
TANK
AI ENGINEER · BACKEND · SURAT, INDIA
I learned this from the bottom up — open-source models, on my own hardware.
Two years building AI that has to work on a Tuesday, not just in a demo.
I ship retrieval systems, agents and guardrails that hold up in production — and the backends underneath them.
YOU KNOW
THE REST
2020 – 2024 · GTU (RNGPIT)
EDUCATION
B.E. Computer Science & Engineering
2020 – 2024
GUJARAT TECHNOLOGICAL UNIVERSITY
(RNGPIT)
Minor specialisation in Artificial Intelligence and Machine Learning.
LLM Engineering — Udemy
HF Agentic AI
HF MCP Course
Python for Data Science — NPTEL
TensorFlow & Keras
9.37CGPA / 10
1stRank, Sem 5
GOLDUniversity Medal, AI/ML
Top 5%NPTEL, topper category
NEKTAR P.L. · JAN – APR 2024
EARLY AI WORK
Where I learned RAG the hard way — from the bottom up, on a laptop.
FEB 2024. MY FIRST REAL TASK.
Make the AI read the documents. And answer from them.
It had a name — RAG. Every tutorial used the OpenAI API, which had just stopped being free.
So I built it from the ground up.
Open-source models, embeddings, chunking, rerankers. Benchmarked FAISS, Milvus and Chroma for latency and recall. All of it had to fit on a 12 GB laptop GPU.
LangChainLlamaIndex
FAISSBGE / GTE
Cross-encoders
The model I needed didn't fit.
Quantization made it fit. Decoder-only vs. seq2seq, then the compression rabbit hole:
GPTQ · AWQ · QAT → GPTQ
IT RAN.
Quantized Zephyr-7B, on a 12 GB laptop GPU
−65%inference memory
+30%answer relevance
1developer on it
WHAT I BUILD WITH
THE STACK
AI & LLMs
GPT-4oClaude
GeminiLlama 4 Maverick
GroqZephyr-7B
Prompt EngineeringGuardrails
GPTQ / AWQ / QATLangChain
LlamaIndexHF Transformers
smolagentsFlowiseAI
n8n
RAG & RETRIEVAL
FAISSMilvus
PineconeChroma
BGEGTE
Hybrid RAGAgentic RAG
RerankersHuman-in-the-loop
FULL-STACK
PythonTypeScript
ReactNext.js
Node.jsFastAPI
FlaskPrisma
PostgreSQLSupabase
NextAuthFrappe / ERPNext
SQL
DEVOPS
VercelFrappe Cloud
DockerNginx
PM2VPS / Linux
GitHub ActionsLet's Encrypt
DR backups
APR 2024 → PRESENT
MY WORK
Intern to engineer. Nine products, shipped.
MOSAIC
US Real EstateHuman-in-the-loop agentic RAG over live property APIs and FAQ documents — and it knows when to stop and ask a human.
LangChainLlamaIndex
FAISSGPT-4o
+25%response accuracy
PAPERWEIGHT
Watlow · MalaysiaClaude reads purchase invoices and files them straight into SharePoint and Excel. Nobody types a number.
n8nClaude API
SharePoint
−80%manual data entry
SVAAYA
Skin & Hair ClinicHIPAA-compliant image analysis, built around what it must never do: give medical advice.
FastAPIGPT-4o
GeminiClaude
HIPAA
4models benchmarked, 1 shipped
CANDORE
IVF ClinicStaged IVF report summaries, written the way a practitioner with 20+ years reads them.
ReactFastAPI
Claude API
−70%report-review time
DECODE AGE
LongevityFrappe back office pulling orders from Amazon, Flipkart, Myntra, Tata Cliq and Shopify — reconciled, then tracked to revenue. Now: manufacturing.
FrappeERPNext
Marketplace APIs
15 hrs/week saved
QUALITY ASIA
ISO Certification BodyClient portal and back office so candidates enrol themselves into ISO certification courses, plus the LMS they sell through.
FrappeFrappe LMS
2portals, one back office
CARDECODE
Workshop SaaSGST-compliant sequential invoicing, an edit-approval workflow and owner analytics. Mine from the first commit.
Next.js 14Prisma
PostgreSQLDocker
−60%invoice prep
~99.5%uptime
PRIME PNEUMATICS
Field ServiceMaintenance visits were tracked on paper. Now: service reports, spare parts, RLS-based RBAC and WhatsApp reminders.
ReactSupabase
RLS
−50%report turnaround
ALSO
Maharana Foods & moreA Frappe back office, a multi-agent concierge, a WhatsApp AI chatbot, and an OCR pipeline at ~96% accuracy.
LlamaIndex WorkflowsFlowiseAI
PaddleOCR
~96%OCR accuracy
NONE OF IT MATTERS IF THE BACKEND IS A HOUSE OF CARDS.
So I got good at the boring parts too — schemas, scheduled jobs, retries, RBAC, and the error handling that decides whether a bad API day becomes an incident.
Turns out the AI was never the hard part.
SIDE QUESTS
My other personal projects.
LIVE
GAIA AGENT
Multimodal agent for the GAIA benchmark — reasoning, web search, page visits and audio transcription.
OSS-120BWhisper
Llama 4 Maverick
TRY THE DEMO →
EVENT PLANNER AI
A CodeAgent on Hugging Face smolagents that plans events through web search and tool automation — with tracing, because agents lie.
smolagentsGroq
LangFuse
SOURCE →
STOCK PRICE PREDICTOR
LSTM and fbprophet forecasting price trends from historical data.
LSTMfbprophet
TensorFlow
SOURCE →
LAPTOP RECOMMENDER
Elastic Net and Random Forest recommending laptops. The one that taught me feature engineering properly.
Elastic NetRandom Forest
Scikit-Learn
SOURCE →
I'M GOING DEEPER INTO
AI ENGINEERING — ON A BACKEND
THAT DOESN'T FLINCH.
Two years in, the thing I'm best at is the seam between them.
END OF ISSUE
Full-stack engineer with AI expertise at RedSoft Solutions, working from Surat, India. If you've got a system that needs to be smart and stay up, I'd like to hear about it.